The invention discloses a high-speed rail bogie sensor fault diagnosis method based on machine learning. The method comprises the following steps: 1) establishing a mapping relation between input parameters and observation parameters, wherein each observation parameter corresponds to a group of input parameters; 2) performing offline training on the neural network by utilizing the historical dataof the observation parameters and the historical data of the corresponding input parameters to obtain a network observer; And 3) carrying out fault real-time diagnosis on the high-speed rail bogie byusing the trained network observer. The method well solves the weaknesses of traditional fault monitoring of the sensor, especially the defects of gradient attenuation, overfitting, local minimizationand the like of a traditional shallow neural network, effectively avoids the complex process of manual feature extraction, and improves the fault diagnosis effect.